Marketing Spending and the Volatility of Revenues and Cash Flows

Transcription

1 Marketing Spending and the Volatility of Revenues and Cash Flows Marc Fischer 1 Hyun Shin 2 Dominique M. Hanssens 3 October 2009 Please address all correspondence to Marc Fischer. 1 Professor of Marketing and Services, University of Passau, Innstr. 27, D Passau, Germany, Phone: +49 (851) , Fax: +49 (851) , marc.fischer@uni-passau.de 2 Assistant Professor of Marketing, College of Management, Long Island University, Brookville, NY 11548, USA, hyun.shin@liu.edu. 3 Bud Knapp Professor of Marketing, UCLA Anderson School of Management, Los Angeles, California , USA, Phone: +1 (310) , Fax +1 (310) , dominique.hanssens@anderson.ucla.edu.

2 1 Marketing Spending and the Volatility of Revenues and Cash Flows ABSTRACT While volatile marketing spending, as opposed to even-level spending, may improve a brand s financial performance, it can also increase the volatility of performance, which is not a desirable outcome. This paper analyzes how revenue and cash-flow volatility are influenced by own and competitive marketing spending volatility, by the level of marketing spending, by the responsiveness of own marketing spending, and by competitive reactivity. From market response theory, we derive predictions about the influence of these variables on revenue and cash-flow volatility. Based on a broad sample of 99 pharmaceutical brands in four clinical categories and four European countries, the authors test for the empirical relevance of our predictions and assess the magnitude of the different sources of marketinginduced performance volatility. The authors find broad support for the expected volatility effects. The results suggest that volatile marketing spending may incur negative financial side effects such as greater financing costs or higher opportunity costs of cash holdings. Thus common volatilityincreasing marketing practices such as advertising pulsing may be effective at the top-line, but could turn out to be ineffective after all costs are taken into account. Keywords: Revenue/Cash-Flow Volatility, Marketing Volatility, Econometric Models, Marketing Metrics

3 2 In order to be effective, marketing executives often deploy their resources in spending bursts, i.e. regimes characterized by on-again, off-again marketing actions, including advertising campaigns, sales promotions and new-product launches. Inasfar as such pulsing causes revenues and cash flows to become more volatile, it may have an unintended negative consequence on their firms financial performance. Indeed, senior executives dislike cashflow volatility since it increases investors perceived uncertainty about future cash flows, incurs additional financing costs for the firm and, accordingly, hurts firm value. Top executives report that they are actively involved in earnings management in order to diminish cash-flow volatility, and are even willing to sacrifice real economic gains (Graham, Harvey, and Rajgopal 2005). Such revenue or cash-flow volatility has traditionally not been of major concern to marketers. However, as marketing moves away from a sole focus on customer-demand impact to a focus on firm-value impact (e.g., Rust et al. 2004; Srinivasan and Hanssens 2009), it adopts an investor perspective, which includes a strong concern about the stability of revenues and cash flows. Marketing spending may indeed be a key source of revenue and cash-flow volatility, which is controllable by the company. For example, Rao and Bharadwaj (2008) show analytically that marketing actions can affect the probability distribution of sales as well as the cash requirements, which in turn influences the distribution of cash flows. Using a single-period model, they argue that the shareholders wealth of the firm can be influenced by such actions. Marketing actions that are routinely carried out by brand managers potentially generate performance volatility. Specifically, how brand managers allocate marketing spending over time may affect cash-flow volatility because it impacts the volatility of brand revenues as well as that of costs. Following the pulsing argument, marketing managers may increase marketing spending

4 3 volatility to improve brand sales, but it may also increase revenue and cash-flow volatility. As long as they are not aware of the potential negative financial effects of such policy, they have no incentive to reduce spending volatility and the resulting cash-flow volatility. Thus, there may be a potential conflict between sales maximization and stable cash-flow generation. In theory, a multi-division firm can diversify its marketing-induced volatility away by strategically coordinating marketing activities across brands and/or regional markets. For example, a beverage manufacturer could ensure that a major campaign for its ice-tea brand does not coincide with another campaign for its bottled-water brand. In practice, however, such coordination is unlikely to occur, because individual brands are treated as profit centers, and regional markets often decide on their own budget allocations. Take, for example, a global consumer packaged goods (CPG) company that markets 100 brands in 50 countries (comparable to P&G) and sets advertising budgets for each quarter of the fiscal year. This requires to coordinate 20,000 spending decisions ( ) per year obviously a challenging task. In fact, an analysis of inter-brand expenditure correlation of manufacturers of our data sample suggests that expenditures are not coordinated across brands, business unit, or regions (see the Appendix A for details). In conclusion, marketing resource allocation at the brand level cannot easily be managed so as to dampen marketing-induced revenue and cash-flow volatility. It is therefore important to study the drivers and consequences of this volatility at the brand level, which is the primary focus of this study. Specifically, this paper addresses three important, yet unanswered questions: Can we make reasonable predictions about how marketing spending affects the volatility of revenues and cash flows based on market response theory? Do we find these effects in practice and how large are they? What are the implications for managerial decision making?

5 4 We will approach these questions as follows. First, we use market response theory to derive the volatility effects of marketing spending. Specifically, we consider the level and the volatility of marketing spending as well as the customer responsiveness to marketing spending as important drivers of revenue and cash-flow volatility. In addition, we derive results on the impact of competitor behavior on volatility. Second, we use a database of 99 pharmaceutical brands from four European countries to test for the expected effects and measure the magnitude of the impact of the various volatility drivers. Third, based on the insights from the conceptual and empirical analysis, we offer recommendations for managers to improve their marketing decisions. The remainder of the article is organized as follows. Following a brief literature review, we develop predictions about the effects of marketing spending on revenue and cash-flow volatility. Next we describe our research methodology to measure the predicted effects. We then present empirical results and discuss the theoretical and managerial implications of our findings. The article concludes with a synthesis of the findings, limitations and suggestions for future research. Prior Research on Marketing-Related Volatility Effects Many marketers believe in the effectiveness of a pulsing strategy, which implies an uneven distribution of expenditures over time. A survey among media-planners in the U.S. reports that almost 70% use pulsing to increase the effectiveness of their spending (Leckenby and Kim 1994). Indeed, theoretical as well as empirical studies have shown that pulsing is a superior strategy that leads to higher revenues and cash flows, provided that market response is subject to threshold effects (e.g., S-shaped response) or differential stimulus effects (e.g., Sasieni 1971;

6 5 Simon 1982; Dubé, Hitsch, and Manchanda 2005). Hence, both marketing theory and practice suggest that marketing managers use pulsing tactics in order to increase revenues and cash flows. However, these tactics may also increase volatility in revenues and cash flows, and thus, create additional financing costs to the firm. In addition to the level of cash flows, which has been the focus in pulsing studies so far, their distribution is also relevant because it determines how much capital must be mobilized over time to finance a specific pulsing strategy. In the Appendix B, we present an example that illustrates the additional financial burden that may arise from a pulsing strategy. The implications for shareholders wealth from volatile marketing expenditures are also analyzed by Rao and Bharadwaj (2008). Only a few empirical studies have addressed the relationship between marketing and revenue/cash-flow volatility to date. Raju (1992) examines the drivers of category sales variability and finds that the magnitude of discounts is positively associated with sales volatility whereas frequency is negatively correlated. By adopting an EGARCH approach, Vakratsas (2007) shows that marketing-mix variables including price, advertising, and distribution affect market-share volatility via the error term. Gruca and Rego (2005) analyze the impact of satisfaction on cash-flow volatility at the firm level, and conclude that satisfied customers are an important asset because they lower the volatility of cash flows. This argument has also been supported by Anderson, Fornell, and Mazvancheryl (2004) and Fornell et al. (2006). While these prior studies provide valuable insights, they do not address the volatility impact on brand revenues and cash flows coming from marketing spending level, spending variation and sales responsiveness. Such a focus is more actionable for marketing managers, but still offers a close link to the company s financial performance.

7 6 Marketing-Induced Volatility Effects We start the discussion of volatility effects with the impact of marketing spending behavior on the volatility of revenues, followed by its effects on the volatility of cash flows. We do this, in turn for a single-firm marketing scenario and for a scenario with competitive actions. Our general argument is that the volatility, the average level and the sales responsiveness of marketing expenditures together affect the volatility of revenues and cash flows. 1 Market Response Theory Our conceptual development is rooted in market response theory, as knowledge of the market response function is important for determining optimal marketing budgets (e.g., Dorfman and Steiner 1954). We assume that (aggregated) brand sales follow a concave relationship with marketing expenditures. By varying conditions such as responsiveness to marketing, we derive predictions about our focal volatility variables. Some of these predictions are straightforward, while others are not. The assumption of a concave response function is justified for several reasons. First, it is theoretically attractive because it allows for diminishing returns which are a prerequisite for marketing budget optimization. Second, it is by far the most frequent type of response function encountered in empirical research. 2 In contrast, empirical evidence for S-shaped aggregate response functions or response functions with differential stimulus variables is rather weak (Hanssens, Parsons, and Schultz 2001, pp ). Since we also do not find evidence for these response types but for a log-log response model in our data, our assumption is consistent with the later empirical analysis. Finally, the results may be generalized to S-shaped and differential stimulus response. Assuming rational behavior, budgets vary within the concave part of these functions.

8 7 Marketing responsiveness is measured by the slope parameter of the response function. It recognizes changing returns to scale and relates actual input (marketing expenditures) in a realistic and straightforward way to actual output (sales). In addition, it is an important parameter for the derivation of optimal marketing budgets (e.g., Dorfman and Steiner 1954) and may easily be converted to elasticity measures that are comparable across units of analysis. We consider two measures of volatility: the range (i.e. the difference between maximum and minimum values) and the variance of marketing expenditures, revenues, and cash flows. Variance is a common measure of variability. Range is a useful metric as well (Alizadeh, Brandt, and Diebold 2002), especially in our brand-level context. Indeed, when managers decide about the temporal distribution of a given marketing budget, they need to determine the extreme points, i.e. when and at what level to set the highest and the lowest spending. We will use a graphical analysis on the range measure to derive volatility effects. For variance, we derive the predictions mathematically. Details are provided in the Appendix C. Single-firm Scenario Effects on Revenue Volatility. Marketing theory and empirical evidence suggest that higher marketing expenditures generally lead to higher sales, but at a decreasing rate (e.g., Hanssens, Parsons, and Schultz 2001). Figure 1 shows this type of market response, where X denotes the level of marketing spending and REV is the level of revenues resulting from a concave response function. Consider first the case of higher spending volatility while keeping the mean spending level constant at X 1. In Figure 1, R 1A (X) and R 1B (X) denote the ranges of expenditures where R 1B (X) is twice as large as R 1A (X). On the ordinate, R 1A (REV) and R 1B (REV) reflect the ranges of revenues associated with these two conditions. Our first

9 8 prediction follows directly from Figure 1: Ceteris paribus, a higher volatility of spending translates into a higher volatility of revenues (prediction I). === Insert Figure 1 about here === Next, consider the effects of a higher mean spending level while holding the range constant. Such a situation can be analyzed in Figure 1 by comparing the revenue realizations associated with expenditure levels around X 1 and X 2. R 1A (REV) and R 2 (REV) are the corresponding ranges of revenues. We conclude that, ceteris paribus, the volatility of revenues decreases with a higher mean level of marketing expenditures (prediction II). The predicted effect of a higher mean spending applies to both a change in spending level for one brand over time (Figure 1 illustrates this case) as well as across brands. Empirically, we expect the effect to come out stronger from a cross-sectional analysis because the cross-sectional variance in spending levels is likely to be higher than the variance over time. In addition, brand differences in spending levels contribute to differences in brand equity and brand loyalty that impact volatility (McAlister, Srinivasan, and Kim 2007; Rego, Billet, and Morgan 2009). Finally, we consider the effect of higher marketing responsiveness on revenue volatility, illustrated in Figure 2. We keep the average level of spending as well as the volatility of spending constant but vary the shape of the response function f(rev). Figure 2 shows that, the more responsive the market is to marketing spending (i.e. f 1 is steeper than f 2 ) the wider the range of revenue realizations. Hence, ceteris paribus, higher marketing responsiveness turns into a larger volatility of revenues (prediction III). === Insert Figure 2 about here === Effects on Cash-Flow Volatility. The results on revenue volatility cannot be automatically applied to cash-flow volatility since an increase (decrease) in revenues is also associated with an

10 9 increase (decrease) in costs. Figure 3 illustrates the cash-flow curve that derives from the market response function of Figure 1. Cash flows are defined as the difference between gross margin (i.e. revenues multiplied by a gross margin ratio) and marketing costs. 3 This curve is concave, left skewed and has a unique maximum reflecting the optimal spend level, as a result of combining a concave response function with a linear cost (marketing budget) function. These properties may lead to results that are different from those of the revenue volatility effects. We focus first on the consequences of a higher volatility of marketing expenditures. Since cash flows are a direct function of marketing expenditures, greater spending volatility causes greater cash-flow volatility. We do not demonstrate this straightforward effect in Figure 3. The corresponding prediction is: Ceteris paribus, a larger volatility of marketing spending translates into a larger volatility of cash flows (prediction IV). === Insert Figure 3 about here === By contrast, the impact of a higher spending level is different from the revenue volatility result. Figure 3 compares three different levels of spending, X 1, X 2 and X 3, while keeping the range of spending constant [i.e. R 1 (X 1 )=R 2 (X 2 )=R 3 (X 3 )]. Obviously, the range of cash flows, Π, decreases with a rise in the spending level when comparing the results for X 1 and X 2 [R 2 (Π)<R 1 (Π)], consistent with the result for revenues. However, the relation turns into the opposite for sufficiently high level of spending, X 3 in Figure 3. The volatility of cash flows is higher compared to X 2 [R 3 (Π)>R 2 (Π)], suggesting a U-shaped effect. The exact inflexion point of the U-shaped curve lies behind the optimal level of spending. 4 Given the well-known flat maximum principle (e.g., Tull et al. 1986) and an asymmetric cash-flow function as depicted in Figure 3, we expect the inflexion point to be rather far behind the optimal spending level for

11 10 many brands. The following prediction reflects our argumentation: Ceteris paribus, the volatility of cash flows follows a U-shape with higher levels of marketing expenditures (prediction V). Finally, higher marketing responsiveness should also lead to a higher volatility of cash flows. This prediction, however, assumes that the average spending level is smaller or equal to the optimal spending level. Under this assumption, the difference between incremental revenues and marketing costs is larger for higher responsiveness making the cash-flow function steeper. The situation reverses for budgets larger than the optimal level when incremental return is negative. Since marketing costs stay the same but incremental revenues are higher due to higher responsiveness the cash-flow function is less steep. Hence, ceteris paribus, higher marketing responsiveness turns into a larger volatility of cash flows, provided the mean level of marketing expenditures does not exceed the optimal level. In our empirical analysis, we expect the effect to be positive because it is unlikely that firms permanently operate in the loss part of the cash-flow function (prediction VI). Our predictions so far are derived from a scenario without competitors (or at least without competitive reactions). We now extend these volatility predictions to account for the impact of competitive actions. Impact of Competition Consistent with economic theory, we assume that firms are profit maximizers and interact with each other. Depending on the number of competitors, the time horizon, the level of information (i.e. the observability of competitive behavior) and the type of competitive interaction (e.g., Nash behavior), an equilibrium in marketing variables such as price or advertising may exist. We do not have evidence that firms engage in expenditure volatility games, in which they set their spending volatility in function of the volatility decisions of their competitors. In fact, it is not

12 11 clear that firms even observe the volatility in marketing spending of their competitors in advance. Thus, we assume that marketing spending volatility is a result of budget setting behavior, not volatility setting behavior under competitive conditions. In the following, we study the impact of competitors marketing spending on the volatility of the focal brand s revenues and cash flows. We do not assume a specific competitive game but rather adopt a reduced-form view that studies observed marketing budgets as a result of competitive interaction. We discuss the effects of competitors' marketing expenditure volatility and the focal firm s reaction behavior on the volatility of its revenues and cash flows. Again, predictions are derived under the assumption that the effects of other variables such as market trend, own marketing mix volatility, etc. are held constant. We do not consider the mean level of competitive spending as a driver of volatility, for lack of a clear understanding how this relates to the volatility of market response for the focal firm. We also do not discuss the effect of a brand s demand responsiveness to competitive activities on the focal volatility variables, because its variation in our dataset is too low to conduct a powerful test of this effect. Competitive-expenditure volatility. The effect of competitive marketing on own brand sales (cross-effect) may be implicitly embedded in the demand function, such as in market-share attraction models, or it can be made explicit by including competitor variables among the predictors. If brand sales is the dependent variable, the cross-effect of competitive expenditures is negative or positive depending on whether the primary demand effect or the substitution effect dominates. Since the cross-effect of competitive spending is structurally equivalent to the effect of own spending, we expect the same effects on volatility of revenues and cash flows as discussed above. This holds true for both directions of the cross-effect, since volatility has no directional meaning. It is therefore straightforward to transfer the results from own spending:

13 12 Ceteris paribus, a larger volatility of competitive marketing expenditures leads to a larger volatility of own revenues and cash flows (prediction VII). As a result, competitors' actions may hurt, not only through their impact on the level of own sales and own cash flows, but also via the volatility effect. How strong this effect is remains an empirical issue. Competitive-reaction behavior. Knowing that competitive actions may impact the level and volatility of their performance, managers are interested in appropriate reactions. We discuss a number of retaliatory actions they may consider to neutralize the adverse effects of competitive actions. We consider only reaction decisions with respect to own marketing expenditures, which include (1) increase own expenditures, (2) decrease own expenditures, or (3) no reaction (e.g., Leeflang and Wittink 1996; Steenkamp et al. 2005). In the first case, the manager tries to synchronize his/her spending behavior with that of his/her competitors, which induces a positive correlation between the expenditures. In reality, reaction occurs with a certain time lag that may lead to divergent correlation structures. With quarterly data as ours, however, this effect vanishes and we should observe a positive correlation if expenditures are synchronized. The effectiveness of the competitive counteraction depends on the relation between the own effect and assumed negative cross-effect, as well as on the magnitude of both expenditures. Provided that the retaliatory action reduces the negative impact of the change in competitive spending, we expect that a higher correlation between own and competitive marketing expenditures reduces the revenue and cash-flow volatilities due to competitive marketing activity. However, if competitive marketing exerts a positive influence on own revenues, the volatility caused by competitive activity will be amplified. As a result, the direction of the effect of a retaliatory measure depends on the sign of the cross-effect. We conclude: Given a negative cross-effect of competitive marketing expenditures, a higher correlation between own and competitive

14 13 marketing expenditures reduces the volatility of own revenues and cash flows. Given a positive cross-effect of competitive marketing expenditures, a higher correlation between own and competitive marketing expenditures increases the volatility of own revenues and cash flows (prediction VIII). In case (2), a negative correlation between own and competitive expenditures results from counter-cyclical spending relative to competition. This does not alter our prediction about the effect of our correlation variable. Finally, the third case (no reaction) introduces zero correlation between the expenditures, and thus no effect on the volatility of revenues and cash flows will occur. Table 1 summarizes our predictions that we now test with a large dataset from the pharmaceutical industry. === Insert Table 1 about here === DATA Data on prescription drugs from two therapeutic areas (cardio-vascular and gastro-intestinal) that cover four product categories are available. Two categories, calcium channel blockers and ACE inhibitors, comprise drugs for the treatment of cardio-vascular diseases. Drugs in the two other categories, H2 antagonists and proton pump inhibitors, are used in gastro-intestinal therapies. These four categories are among the largest prescription-drug categories. They differ in their therapeutic principles to treat diseases like hypertension or acid related gastro-intestinal disorders. Data, collected by IMS Health, are available on a quarterly basis for a time period of 10 years ( ) covering the growth and maturity phases of the analyzed categories. They include unit sales (normalized over different application forms of the drug and transformed into

15 14 daily dosages by a brand-specific dosage factor), revenues, and aggregate marketing expenditures on detailing, journal advertising and other communications media. Detailing has the lion s share in expenditures with more than 90%. Monetary values are in 1996 US$ and have been deflated by country-specific consumption price indexes. The data cover four European countries: France, Germany, Italy, and the UK, and comprise sixteen product markets (4 categories 4 countries). We analyze data on 99 brands, which were marketed by 26 pharmaceutical firms. Table 2 shows the descriptive statistics of the variables used in the estimation equations. Revenues average about $9.2 million per quarter, cash flows are about $5.0 million, and average marketing spending amounts to about $1.0 million. There is also considerable variation in the data across brands and time, as indicated by the standard deviations and the volatility measures in Table 2. Volatility is particularly high with respect to marketing spending. Moving variance is about $151.1 million (or $0.4 million in terms of standard deviation) and moving range is about $0.8 million, virtually as high as the mean spending. Plots of marketing spending over time (not shown) reveal substantial volatility for many brands in our sample. This volatility also translates into substantial volatility at the portfolio level (see the Appendix A, Table A.1 again). === Insert Table 2 about here === METHODOLOGY The empirical testing of the expected effects on the volatility of revenues and cash flows proceeds in two steps. The first step estimates a market response model that relates brand unit sales to relevant variables, among them own and competitive marketing expenditures. We also test whether market response is subject to threshold or differential stimulus effects that would

16 15 justify pulsing as an optimal spending strategy. This market response model provides us with brand-specific estimates of marketing responsiveness. In a second step, these responsiveness estimates are used as predictor variables in a model that explains differences in volatility of revenues and cash flows (volatility models). 5 In addition, we use the results of the market response model to remove the effects of exogenous factors such as seasonality and trend from the brand sales time series. Such factors are outside the control of management and are therefore not relevant for the study of marketing spending impact on volatility. Brand expenditures are not subject to trend or seasonality as specification tests revealed. Market Response Model We apply the multiplicative interaction model, a standard response model, to explain brand unit sales. This functional form has received large empirical support, has been found useful in normative applications, and incorporates interaction effects in a parsimonious way (e.g., Hanssens, Parsons, and Schultz 2001). To allow for pulsing as a superior strategy we augment the model with a differential-stimulus effect (Simon 1982) that captures any extra demand lift due to pulsing (see the Appendix D for the exact specification). We include variables that are relevant to the international markets over the ten-year sample period. Specifically, the response model incorporates own and competitive marketing expenditures on detailing, professional journal advertising, and direct marketing, including lagged effects. More than 90% of expenditures are allocated to detailing. We account for brand heterogeneity in demand, e.g., quality, brand equity, order of entry, by estimating brand-specific fixed effects. The model further includes three types of covariates, i.e., seasonal dummies, a trend variable (elapsed time since launch of a brand) to control for life-cycle effects, and a country s gross domestic product (GDP) as a proxy of the overall economic condition of a

17 16 country. Distribution and price are not relevant variables in our context. In the European countries covered by our data, pharmacies are required to list every approved drug, resulting in 100% distribution for the drugs in our sample. Prices were highly regulated during the observation period and therefore not used as a tactical marketing instrument. Details on model estimation are provided in the Appendix D. Volatility Models Structural Equations. Let V(REV) denote the volatility of revenues measured in terms of variance or range, respectively, V(MKT) represent the volatility of own marketing expenditures, A(MKT) be the average level of own marketing expenditures, V(CMKT) denote the volatility of competitive marketing expenditures, CORR represent the correlation between own and competitive marketing expenditures, RESP denote total marketing responsiveness (including lagged and current effects), X denote a vector including the remaining variables of the brand sales model as specified in Equation (A.11) in the Appendix D (i.e., brand-fixed effects to control for order of entry, quality, etc., trend, seasonality, and GDP as surrogate for general demand), γ be a parameter vector to be estimated, and ν be an error term with variance ξ. Omitting brand and time subscripts for the moment, we specify the revenue volatility model as follows: γ1 γ2 γ3 = γ ( γ + γ 5 + Xγ + ν) with ν N ( 0, ξ). V REV V MKT A MKT V CMKT Exp CORR RESP 0 4, (1) We assume the relationship between revenue volatility and its drivers to be multiplicative. Thus the variables interact with each other, consistent with the results from the market response theory discussion. The correlation between own and competitive marketing expenditures and the estimated marketing responsiveness parameter appear as part of an

18 17 exponential function because they may become negative. 6 The parameters γ 1-3 can be directly interpreted as elasticities and facilitate the comparison of volatility drivers. We subsequently describe how we transform the dataset to remove the X-variables, which are not the focus in this study. Since cash flows are constructed from revenues and costs, revenue volatility enters the cash-flow volatility equation: δ1 δ2 δ3 0 4 with υ N ( 0, ψ), V CF = δ V REV V MKT A MKT Exp δ A MKT + υ,, (2) where V(CF) denotes the volatility of cash flows, δ is a parameter vector to be estimated, and υ represents an error term with variance ψ. The effects of competitive-marketing-expenditure volatility, competitive reaction, marketing responsiveness, and X-variables on cash-flow volatility are mediated through revenue volatility. In addition, revenue volatility mediates the impact of own expenditures. Since own expenditures also enter the cash-flow equation as cost we expect an additional direct effect on cash-flow volatility. Finally, note that specification (2) allows for a U-shaped influence of the level of marketing expenditures on cash-flow volatility, consistent with our prediction V. This situation occurs if δ 3 <0 and δ 4 >0. We further allow the error terms to be correlated across the two equations. Data Transformation. By using the estimates of the brand sales model, we remove the effects of exogenous market factors such as seasonality, trend, and overall economic condition (measured by the GDP), and derive an adjusted unit-sales time-series for each brand. We multiply the unit sales with the brand's unit price and arrive at adjusted brand revenues. We then multiply the adjusted revenues by a cash contribution margin of 65% as given in Myers and

19 18 Howe (1997). From these gross cash flows we subtract the marketing expenditures and arrive at the final variable of adjusted brand cash flows. 7 The volatility of the adjusted revenues and cash flows is measured by the variance or range of these quantities over a time period of 8 quarters. Consequently, we use the first two available years of sales for each brand as an initialization period. We compute the volatility measure of the subsequent period by dropping the first period and including the information of the following period. We continue until the end of the brand-specific time series and thus obtain a time series of moving volatility measures of adjusted revenues and cash flows (movingwindow analysis). This procedure is also applied to compute moving volatilities for own and competitive marketing expenditures and the moving average of own marketing expenditures. We denote moving volatilities with MV and moving averages with MA. The application of moving-window analysis is well established in the accounting literature (e.g., Kothari 2001) and is justified for two reasons. First, it increases sample size and therefore improves the power of statistical tests. Note that observations are inevitably lost due to the calculation of the volatility measures. Second, it accounts for possible dynamic effects. Capital markets research has shown that it often takes some time until economic effects have fully materialized in earnings volatility. Estimation Equations. The use of moving windows is helpful to increase the power of statistical tests due to the increase in degrees of freedom, but it is also likely to generate serially correlated errors in the time series of adjusted revenues and cash flows. We therefore transform expressions (1) and (2) into a series of relative differences. By taking the total differentials of the log-transformed equations (1) and (2), we obtain (see the Appendix E for details):

20 19 ΔMV AREV ΔMV MKT ΔMA MKT ΔMV CMKT ikst = γ1 ikst + γ2 ikst + γ3 ikst ikst 1 ikst 1 ikst 1 ikst 1 γ4 Δ MA CORR +Δνikst, ikst MV AREV MV MKT MA MKT MV CMKT + (3) ΔMV ACF ΔMV AREV ΔMV MKT ΔMA MKT ikst = δ1 ikst + δ2 ikst + δ3 ikst ikst 1 ikst 1 ikst 1 ikst 1 MV ACF MV AREV MV MKT MA MKT δ Δ MA MKT +Δ 4 υikst, ikst + (4) where, MV(AREV) ikst = Moving volatility of adjusted revenues of brand i in therapeutic area k, country s and period t MV(MKT) ikst = Moving volatility of marketing expenditures of brand i in therapeutic area k, country s and period t MA(MKT) ikst = Moving average of marketing expenditures of brand i in therapeutic area k, country s and period t MV(CMKT) ikst = Moving volatility of marketing expenditures of brand i s competitors in therapeutic area k, country s and period t MA(CORR) ikst = Moving average correlation between own and competitive marketing expenditures of brand i in therapeutic area k, country s and period t. MV(ACF) ikst = Moving volatility of adjusted cash flows of brand i in therapeutic area k, country s and period t Δ = First-difference operator. Equations (3) and (4) represent the original equations (1) and (2) in terms of relative differences. Unlike absolute differences, this representation not only reduces serial correlation, but also controls for brand-size effects. For example, bigger brands are expected to have larger absolute changes in revenues, cash flows and marketing spending. Equations (3) and (4) establish an equation system with possibly correlated errors across equations. Revenue volatility is the only endogenous variable occurring at the right hand side of Equation (4). Thus, the system is recursive and Generalized Least Squares (GLS) provides efficient estimates (Zellner 1962). Since first-differencing may not completely remove serial correlation we also allow for equation-specific autocorrelation coefficients in the variancecovariance matrix.

21 20 The first-differencing procedure eliminates the time-invariant marketing responsiveness variable that is part of the revenue volatility model (1). 8 To measure its influence, we linearize (1) first via log-transformation and then build a cross-sectional regression model by obtaining averages of all time-varying variables. The resulting equation can be estimated with Ordinary Least Squares (OLS). However, the marketing-responsiveness parameters of the first stage are measured with sampling error that vanishes in the limit. As a consequence, OLS estimates from the second stage regression will be consistent but their standard errors may be biased (Murphy and Topel 1985). Following Nijs, Srinivasan, and Pauwels (2007), we obtain corrected standard errors by a bootstrapping procedure with 10,000 replications. RESULTS Brand Sales Model We estimate the multiplicative brand sales model with and without the differential stimulus (DS) effect (see Equation A.11 in the Appendix D again). The DS-effect is insignificant in all markets where estimation was feasible, and information criteria such as the Schwartz Information Criterion (SC) disfavor the models with DS-effect. Alternatively, we estimate an S-shaped response model (details are available from the authors). This specification also turned out to be inferior compared to the standard log-log model. Hence, we conclude that market response follows a diminishing-returns pattern in the analyzed markets, which does not justify a pulsing policy. All following results are thus based on the log-log model without DS-effect. The multiplicative brand sales model describes sales evolution in the markets very well. The average total marketing elasticity, weighted by relative standard errors to account for estimation uncertainty, is.192, which is well in line with recently reported results (e.g., Fischer

22 21 and Albers 2009). The impact of competitive marketing activities is negative, with a mean value of In general, there is substantial variation in the marketing impact estimates, which we use as a predictor in our volatility models. Recall that we use the total effect which is the sum of current and lagged marketing elasticity. Volatility Models Table 3 shows the estimation results for the revenue and cash-flow volatility models by using either (adjusted) variance or range as dependent variable. Our focal predictor variables explain a substantial part of variance in observed (i.e. unadjusted) revenue and cash-flow volatility in estimation and holdout samples underlining the relevance of marketing activities for performance volatility. To form holdout samples we excluded the last four quarters (20% of total cases) in the first-difference models and the last 20 brands (20% of cases) in the crosssectional model. === Insert Table 3 about here === In the following discussion we focus on variance as volatility measure and on the results from first-difference models. Since the effect of marketing responsiveness, which does not vary within but across brands, can only be estimated by a cross-sectional model we also report on the results of the cross-sectional regression model. However, due to the missing time variation and the substantially lower number of observations in this model, the effects for the time-varying variables should be interpreted with caution. We discuss first estimates from the revenue volatility model and then turn to the cashflow volatility model. The volatility of marketing expenditures measured by their variance increases the volatility of revenues and supports our first prediction, with an estimated elasticity of.273 (p<.05). 9

23 22 The first-difference model also supports our second prediction on the influence of the level of marketing expenditures on revenue volatility; but the coefficient is not significant at p<.05. We obtain a significant negative effect from the cross-sectional regression (-1.99, p<.05). Note that this variable has been divided by average brand unit sales in order to control for brandsize effects. As expected, the effect comes out stronger in a pure cross-sectional regression. Marketing responsiveness drives revenue volatility (8.11, p<.05), supporting our third prediction. The associated elasticity of.811 (= ) is substantial. We do not find significant effects (p>.05) for the volatility of competitive marketing expenditures in the firstdifference regression and the cross-sectional regression. The correlation of own and competitive marketing expenditures shows a significant negative effect on revenue volatility (-.262, p<.05). Since the effect of competition is predominantly substitutive in our data, the sign of the effect is consistent with our expectation. As expected, revenue volatility is an important driver of cash-flow volatility, with an elasticity of 1.36 (p<.05). Its lower boundary value is the squared profit margin, which would be achieved if cash flows consisted only of revenues multiplied by the profit margin. The direct effect of the volatility of marketing expenditures is positive and significant, with a value of.535 (p<.05). This coefficient represents the volatility effect due to the cost component of marketing expenditures. In order to fully evaluate the predicted effect of expenditure volatility on cashflow volatility, we need to consider the total effect. Table 4 displays the total effects in terms of elasticity, which facilitates the interpretation and comparison of the magnitude of effects. The total effect of expenditure volatility on cashflow volatility amounts to.906 (= ; p<.05). Hence, we find strong support for our fourth prediction. Interestingly, this elasticity is more than three times higher than for revenue

24 23 volatility. We also find strong support for the expected U-shaped influence of the level of marketing expenditures on cash flows (-2.38, p<.05 and.003, p<.05; see table 4). The direction of the influence of marketing responsiveness on cash-flow volatility is also consistent with our prediction. Its elasticity is high with a value of 1.10 (p<.05). The volatility effect of the volatility of competitive marketing expenditures is not significant, which may be due to the fact that the estimated cross-effects are rather small and not uniform in sign across all categories. We find, however, support for the expected cash-flow volatility effect of the correlation of own and competitive marketing expenditures though the associated elasticity is small (-.003; p<.05). === Insert Table 4 about here === Both tables 3 and 4 show also the results for models when we take range instead of variance as volatility measure. Overall, the results are consistent with the results using variance as volatility measure. Robustness of Findings We performed several analyses to verify the robustness of these results. First, we varied the window of the volatility measures. Instead of 8 quarters we computed volatility measures based on 4 quarters and 12 quarters. The results were similar but model fit deteriorated, underlining that the 8-quarter window is the best choice for our dataset. Second, we created volatility variables that do not overlap over time periods. For example, the first observation of an 8- quarter-based variance variable includes the first 8 quarters, the second observation is based on the subsequent 8 quarters, and so forth. This procedure reduces the sample size to only 292 observations. The results did not change materially, though the standard errors increased. Third, we estimated a linear model and a semi-log model in addition to the already reported brand slaes models. The results of the Davidson-McKinnon comparative test (Greene 2004) support our log-

25 24 log specification. Fourth, we verified whether the results are influenced by collinearity. The condition indices of the models were well below the critical value of 30 (Greene 2004). DISCUSSION Managerial Implications Our study provides insights that invite marketing decision makers to think differently about the consequences of their actions. First, our results show that higher marketing spending volatility leads to a higher volatility of revenues as well as cash flows. Thus managers who decide on the timing of media plans, promotion plans, product launches etc. can influence the volatility of both the top-line and bottom-line performance of their brands. Since marketing costs grow linearly while revenues grow at a decreasing rate, their impact on cash-flow volatility is larger than on revenue volatility. Second, stronger market response parameters also translate into higher volatility of revenues and cash flows. Thus, on the one hand, larger response parameters are good news for the marketing manager because his/her expenditures produce higher returns. On the other hand, higher responsiveness has a dark side since it makes revenues and cash flows more volatile, even if spending volatility itself does not change. Third, we find that a higher mean level of marketing expenditures reduces revenue volatility, holding spending volatility constant. Higher spending also decreases the cash-flow volatility for typical left-skewed cashflow distributions up to a certain level. That level is greater than the optimal spending level. Some marketing tactics, such as advertising campaigns, are used frequently and involve a volatile deployment of the marketing budget. Sometimes these tactics improve a brand s top-line results, sometimes they do not, but in either case, we expect them to have an effect on the volatility of both revenues and cash flows, as our conceptual and empirical analysis suggest.

26 25 Since volatility incurs additional financial costs, even revenue-effective volatile marketing tactics may turn out to be harmful to the bottom line. This creates a managerial tradeoff, as illustrated in Figure 4. The matrix provides brand managers with a means to evaluate their marketing policy. If the effect of marketing volatility on the level of revenues/cash flows is small or non-existent, there is no need to increase marketing volatility, and in fact it should be avoided. If the effect on sales is high, managers need to find the right balance between that positive impact and its negative financial side effect. === Insert Figure 4 about here === The volatility of brand expenditures in our data was quite high. In particular, we observed several new brand entries for which firms adopted highly volatile spending behavior in the first two years after launch, a time when marketing spending often exceeds brand revenue. In contrast to that, we do not find evidence for a market response that justifies pulsing. As a consequence, managers are in the double jeopardy scenario of Figure 4; and there is no economic reason to deploy volatile marketing tactics as it only adds financing costs to the losses that are already being incurred. 10 For the newly introduced brands in our observation period (61 brands out of 99), we estimate these costs to amount to US$ 5.1 million on average (median = 1.8 million) in the first two years after launch. Similarly, different brands have different levels of marketing spending and our results show that those with higher spending levels enjoy protection against performance volatility, especially cash-flow volatility, so long as their expenditures are economically reasonable, i.e. they are not too far beyond the Dorfman-Steiner optimal levels. On the other hand, higher spending quality (as assessed by responsiveness) comes at a cost of increased performance

27 26 volatility in connection with volatile spending behavior, and in that sense good marketing and stable business performance are difficult to reconcile. Research Implications Our findings contribute to the advancement of knowledge in marketing. Cash-flow volatility has been overlooked in marketing for a long time (Srivastava, Shervani, and Fahey 1998). While recent research (e.g. Gruca and Rego 2005; Rao and Bharadwaj 2008) has focused on marketing s potential to reduce volatility and its associated financial side effects at the firm level, our study is the first to describe its potential to increase volatility. We do so by relying on extant market response theory, which allows us to make the formal connection between marketing spending, marketing responsiveness and revenue and cash-flow volatility. The empirical application of our theoretical framework to a large dataset from the pharmaceutical industry creates interesting findings. We learn that the volatility of marketing expenditures has a sizeable effect on revenue volatility, and its effect on cash-flow volatility is even higher. Marketing responsiveness has a strong effect on revenue volatility, which translates into a substantial impact on cash-flow volatility. Finally, our analysis shows that higher marketing-spending levels help reduce the volatility of revenues and cash flows. This conclusion is similar to the firm-level finding of McAlister, Srinivasan, and Kim (2007) that marketing spending reduces the systematic risk of the firm. Thus while generous marketing spending may be favorable for the firm in the long run, at the same time, allocating these budgets across brands in a pulsing pattern may be less favorable. Revenue and cash-flow volatility may also arise from other marketing-mix activities such as promotions and new-product introductions. The extant promotion literature shows that promotions lead to higher revenues, at least in the short run (e.g., Blattberg, Briesch, and Fox